Abstract:Artificial intelligence (AI) in the form of deep learning bears promise for drug discovery and chemical biology, $\textit{e.g.}$, to predict protein structure and molecular bioactivity, plan organic synthesis, and design molecules $\textit{de novo}$. While most of the deep learning efforts in drug discovery have focused on ligand-based approaches, structure-based drug discovery has the potential to tackle unsolved challenges, such as affinity prediction for unexplored protein targets, binding-mechanism elucidation, and the rationalization of related chemical kinetic properties. Advances in deep learning methodologies and the availability of accurate predictions for protein tertiary structure advocate for a $\textit{renaissance}$ in structure-based approaches for drug discovery guided by AI. This review summarizes the most prominent algorithmic concepts in structure-based deep learning for drug discovery, and forecasts opportunities, applications, and challenges ahead.

What problem does this paper attempt to address?

### Problems Addressed by the Paper The paper "Structure-Based Drug Discovery and Deep Learning" aims to explore how deep learning techniques can accelerate structure-based drug discovery (SBDD). Specifically, the paper focuses on the following key issues: 1. **Affinity Prediction**: How to use structural information of proteins and ligands to predict their binding affinity. 2. **Binding Site Detection**: How to identify "druggable" binding sites in proteins that can serve as targets for drug design. 3. **De Novo Design**: How to design new molecules with specific biological activities based on the information of target proteins. ### Background Traditional ligand-based methods have achieved some success in drug discovery, but structure-based methods have greater potential to address unresolved challenges, such as affinity prediction for new protein targets, elucidation of binding mechanisms, and rationalization of related chemical kinetic properties. In recent years, the development of deep learning methods and the improved accuracy of protein tertiary structure prediction have provided new opportunities for structure-based drug discovery. ### Main Tasks 1. **Drug-Target Interaction Prediction**: - **Sequence-Based Methods**: Use amino acid sequences and ligand representations to predict interactions. - **3D Structure-Based Methods**: Train using atomic coordinates of protein-ligand complexes. - **Surface-Based Methods**: Represent binding pockets and ligands as surfaces for binding affinity prediction. 2. **Binding Site Detection**: - **Sequence-Based Models**: Predict which amino acid residues are involved in ligand binding. - **3D Structure-Based Models**: Use spatial information of proteins to detect potential binding sites. - **Surface-Based Models**: Represent protein structures as continuous shapes, combining geometric and physicochemical features for prediction. 3. **De Novo Design**: - **Sequence-Based Methods**: Transform the problem into a machine translation task to generate high-affinity protein-ligand pairs. - **3D Structure-Based Methods**: Generate new 3D molecular graphs or string-form molecules based on the tertiary structure information of proteins. ### Challenges and Future Directions Despite some progress in structure-based drug discovery, many challenges remain, such as the detection of shallow binding sites, identification of sub-pockets, and prediction of allosteric binding sites. Future research directions include increasing pocket coverage, improving model generalization, and developing more efficient generative models. ### Summary By reviewing existing deep learning methods, this paper demonstrates the great potential of structure-based drug discovery and points out the limitations and future directions of current research. By leveraging deep learning techniques, it is possible to explore the vast chemical space more effectively and accelerate the drug discovery process.

Structure-based drug discovery with deep learning

Artificial intelligence in drug discovery: recent advances and future perspectives

Concepts of Artificial Intelligence for Computer-Assisted Drug Discovery

Deep Learning in Structure-Based Drug Design

De Novo Structure-Based Drug Design Using Deep Learning

Structure-based drug design with geometric deep learning

Recent Progress of Deep Learning in Drug Discovery

The rise of deep learning in drug discovery

Deep learning for low-data drug discovery: hurdles and opportunities

Artificial intelligence to deep learning: machine intelligence approach for drug discovery

Application of Deep Learning Neural Networks in Computer-aided Drug Discovery: A Review

Deep Learning in Drug Discovery: Current Landscape and Future Prospects

A compact review of progress and prospects of deep learning in drug discovery

Data Integration Using Advances in Machine Learning in Drug Discovery and Molecular Biology

Deep Learning in Drug Design—Progress, Methods, and Challenges

Advances in Deep Learning Assisted Drug Discovery Methods: A Self-review

Deep learning tools for advancing drug discovery and development

Advances in De Novo Drug Design: From Conventional to Machine Learning Methods

Machine learning in drug design: Use of artificial intelligence to explore the chemical structure–biological activity relationship

Exploring the Artificial Intelligence and Machine Learning Models in the Context of Drug Design Difficulties and Future Potential for the Pharmaceutical Sectors

Recent applications of deep learning and machine intelligence on in silico drug discovery: methods, tools and databases